← blog  /  scrapers

Workable, Recruitee, Personio and BambooHR job board APIs: no key, and a date that costs a request

2026-11-16  ·  12 min read  ·  scrapers

Workable, Recruitee, Personio and BambooHR give a board's open jobs with no key, whole in one request; only BambooHR's list has no date. The four addresses, the catch in each, and about sixty lines of Python that put all four into one shape.

Not every company writes its own careers page. Many small and mid-size ones rent a page from a hiring system (an ATS), and four common ones are Workable, Recruitee, Personio and BambooHR. Each has a public address that lists a company's open jobs, and none of the four asked us for a key, a login or a cookie. Our earlier post on four other job board APIs named these four as the systems it left out.

We read twelve public boards, three of each system, on 9 November 2026 at 01:52 UTC (03:52 in Jerusalem): Hugging Face, United Motors Group and Kobiton on Workable; Miaplaza, UpSlide and KP Snacks on Recruitee; ottonova, pro4all and fact-finder on Personio; scribd, californiaclosets and mcgregoreba on BambooHR. The robots.txt of each of the ten hosts came first and leaves the paths we used open. Then came 29 requests, 1.5 seconds apart, and one more four minutes later. No host walled us. Every number below is from that run, and days are counted in UTC.

The short version#

  • Workable: no key; the whole board in one request (6, 29 and 0 jobs); an exact day, no time; no pay field on any of the 35 jobs.
  • Recruitee: no key; the whole board in one request (6, 4 and 21 offers, 620 KB for the biggest); date and time to the second, with "UTC" typed after it; a pay block on every offer, filled on 9 of 31, its numbers written as text.
  • Personio: no key; XML instead of JSON; the whole board in one request (8, 4 and 4 positions); date and time to the second; no pay tag on any of the 16 positions.
  • BambooHR: no key; the whole list in one request (2, 9 and 0 jobs); no date in the list, so one more request per job; pay only as free text, on that same second request.

None of the four paged. The Workable and Recruitee answers have no next, total or offset at the top, and ?limit=2 changed nothing on either: 6 jobs and 6 offers came back. The biggest board we read had 29 jobs, so what a board of 500 does is left open.

Where the jobs are#

Where the jobs aretext
Workable   GET  https://apply.workable.com/api/v1/widget/accounts/{account}
Recruitee  GET  https://{company}.recruitee.com/api/offers/
Personio   GET  https://{company}.jobs.personio.com/xml
BambooHR   GET  https://{company}.bamboohr.com/careers/list
BambooHR   GET  https://{company}.bamboohr.com/careers/{id}/detail

The {account} or {company} part is the name in the company's careers link. Personio ends in .com or .de: take the ending from the link.

The name in a careers linktext
apply.workable.com/huggingface   huggingface
miaplaza.recruitee.com           miaplaza
pro4all.jobs.personio.com        pro4all
scribd.bamboohr.com/careers      scribd

Workable#

One Workable boardbash
curl "https://apply.workable.com/api/v1/widget/accounts/huggingface"

The answer is an object: the company's name, its description as HTML, and jobs. Hugging Face gave 6 jobs in 4.2 KB, United Motors Group 29 in 18.7 KB, and Kobiton an empty list in 1.0 KB, still HTTP 200. A job has 19 keys, among them title, shortcode, url, published_on, created_at, city, state, country, telecommuting, employment_type, department and a locations list. No key on any of the 35 jobs names a salary, pay or compensation.

Three catches:

  • The job description is a parameter away. The list has none. With ?details=true the six Hugging Face jobs weighed 36.6 KB instead of 4.2 KB, and description was the only new key.
  • Two dates, and neither orders the list. published_on and created_at are days with no time, like 2026-09-21. They differ on 19 of the 35 jobs, and published_on is the later one every time: Hugging Face's oldest created_at is 2022-10-19, its oldest published_on 2024-10-17. On both boards the list was in the order of neither date. The code below uses published_on.
  • A place can be only a country. country was there on all 35 jobs, but city and state are missing on 3 of Hugging Face's 6, and the locations list has a hidden flag, true on 3 of those 6. telecommuting is a real flag: true on all 6 Hugging Face jobs, false on all 29 of United Motors Group's. employment_type was empty or missing on 5 of the 35.

Recruitee#

One Recruitee sitebash
curl "https://miaplaza.recruitee.com/api/offers/"

The answer is {"offers": [...]}. Miaplaza gave 6 offers in 236.9 KB, UpSlide 4 in 167.2 KB and KP Snacks 21 in 619.7 KB. An offer has 56 keys, and the status of all 31 was published.

Three catches:

  • It is heavy, and you cannot ask for less. The first Miaplaza offer alone is 35,479 bytes written out as JSON: translations 12,785, open_questions (the application form) 7,885, description 6,777 and requirements 5,485. The description is in translations a second time, equal on all 31 offers. ?limit=2 still gave all 6.
  • Pay is a block on every offer, and mostly empty. salary is on all 31 offers, with min, max, currency and period. A min or max is filled on 9: all 6 at Miaplaza (per hour, in USD), 1 of 4 at UpSlide and 2 of 21 at KP Snacks. The values are text, "30285.60" and not 30285.6, and period was hour, year or month.
  • A remote job's place says "Remote job". location is a text such as "Paris, Île-de-France, France", but on all 6 Miaplaza offers it says "Remote job" and city says "Fully Remote". A locations list of more than one entry was on 2 of the 31 offers, while location is a single text. The three flags remote, hybrid and on_site were true on 6, 6 and 19 of the 31.

The date is published_at, written "2026-11-07 00:37:39 UTC". Python 3.10's fromisoformat refuses that, so the code takes the first 10 characters. The list was newest first on one board and in no date order on the other two.

Personio#

One Personio feedbash
curl "https://pro4all.jobs.personio.com/xml"

The answer is XML: a workzag-jobs element with one position for each job. ottonova gave 8 positions in 98.2 KB, pro4all 4 in 2.6 KB and fact-finder 4 in 3.4 KB. The positions of a board use 15 or 16 different tags, among them id, name, office, department, employmentType, schedule, createdAt and jobDescriptions. None of them is pay. createdAt is a time with an offset, 2026-03-23T15:24:29+00:00, which Python reads as it is. The lists were in date order on none of the 3 boards.

Three catches:

  • There is no link in it. The code builds one from the id, https://{host}/job/{id}. A follow-up request four minutes later opened the first ottonova one: HTTP 200, 99,284 bytes, with the job's name in the page. ottonova's board is on .de, the other two on .com.
  • The description may be missing. ottonova's feed had one for 8 of 8 positions: 42 parts in all, HTML inside CDATA, 34 of them with tags. On the other two boards the jobDescriptions element was empty on 8 of 8 positions, so there the feed has no text at all.
  • Offices are text, and there can be several. office is a city in the company's own language, "München" at ottonova and "Munich" at fact-finder, and additionalOffices lists more: 9 of the 16 positions had more than one. The offices of pro4all included "Hybrid", which is not a place.

BambooHR#

One BambooHR list, then one jobbash
curl "https://californiaclosets.bamboohr.com/careers/list"
curl "https://californiaclosets.bamboohr.com/careers/27/detail"

The list answer has two keys, meta and result. meta holds totalCount, and it matched the number of rows on all 3 boards. scribd gave 2 jobs in 0.7 KB, californiaclosets 9 in 3.0 KB, and mcgregoreba an empty list in 37 bytes. A job has 10 keys, among them id, jobOpeningName, departmentLabel, employmentStatusLabel, location, atsLocation, isRemote and locationType. There is no date, no description and no pay in it.

Three catches:

  • The date is one request away, for each job. The detail address has datePosted, a day such as 2023-09-28, and the description. It weighed 18,424 and 17,249 bytes for two californiaclosets jobs, against 3.0 KB for the whole list of 9. We read the first and last job of each list, and 4 of 4 had a date. californiaclosets' first job was posted 1,138 days before the run and its last 25 days before; the two scribd jobs were posted in 2019, 2,618 and 2,551 days before, and both are still "Open".
  • Pay is free text. compensation on that same address was "$50,000-$80,000 per year" and "$24-$27/hr" for the two californiaclosets jobs, and empty for both scribd jobs: 2 of 4 reads. The link the code builds, /careers/{id}, was the jobOpeningShareUrl on 4 of 4.
  • The place may be in another object. On 2 of californiaclosets' 9 jobs (Cabinet Installer in Birmingham and in Huntsville) location has no city, and the place is in atsLocation, which holds the country as well; the code falls back to it. Nothing says remote: isRemote is null on all 11 jobs, and locationType is "0", "1" or "2" (5, 2 and 2 at californiaclosets) with no key to what they mean.

One shape for all four#

The four answers disagree on how a date is written (a day, a time with UTC after it, a time with an offset, or none), on where the place lives, and on whether "remote" is a flag. This puts them into the row shape of the earlier post. It uses only the Python standard library and ran as printed over the answers of the run above.

Read four more job boards into one row shapepython
import json, urllib.request
import xml.etree.ElementTree as ET

UA = {"User-Agent": "voiddo-guide/1.0 (+https://voiddo.com/scrapers)",
      "Accept": "application/json, application/xml;q=0.9"}


def get(url):
    with urllib.request.urlopen(urllib.request.Request(url, headers=UA), timeout=30) as r:
        return r.read().decode("utf-8")


def row(ats, company, title, location, posted, url, remote=None):
    return dict(ats=ats, company=company, title=title, location=location,
                remote=remote, posted=posted, url=url)


def place(*parts):  # "Paris, Île-de-France, France", without the empty parts
    return ", ".join(p for p in parts if p)


def workable(account):
    jobs = json.loads(get(f"https://apply.workable.com/api/v1/widget/accounts/{account}"))["jobs"]
    return [row("workable", account, j["title"], place(j.get("city"), j.get("state"), j.get("country")),
                j["published_on"], j["url"], j.get("telecommuting")) for j in jobs]


def recruitee(company):
    offers = json.loads(get(f"https://{company}.recruitee.com/api/offers/"))["offers"]
    return [row("recruitee", company, o["title"], o.get("location"), o["published_at"][:10],
                o["careers_url"], o.get("remote")) for o in offers]


def personio(company, tld="com"):  # the feed is XML, and it holds no link: the link is built from the id
    host = f"{company}.jobs.personio.{tld}"
    out = []
    for p in ET.fromstring(get(f"https://{host}/xml")).findall("position"):
        offices = [p.findtext("office")] + [o.text for o in p.findall("additionalOffices/office")]
        out.append(row("personio", company, p.findtext("name"), "; ".join(o for o in offices if o),
                       p.findtext("createdAt")[:10], f"https://{host}/job/{p.findtext('id')}"))
    return out


def bamboohr(company):  # the list has no date: bamboohr_date() below costs one more request for each job
    host = f"{company}.bamboohr.com"
    out = []
    for j in json.loads(get(f"https://{host}/careers/list"))["result"]:
        where = j.get("location") or {}
        if not where.get("city"):  # on some jobs the place is in atsLocation instead
            where = j.get("atsLocation") or {}
        out.append(row("bamboohr", company, j["jobOpeningName"], place(where.get("city"), where.get("state")),
                       None, f"https://{host}/careers/{j['id']}", j.get("isRemote")))
    return out


def bamboohr_date(company, job_id):
    detail = json.loads(get(f"https://{company}.bamboohr.com/careers/{job_id}/detail"))
    return detail["result"]["jobOpening"]["datePosted"]
Run itpython
BOARDS = [(workable, "huggingface"), (workable, "umg-careers"), (workable, "kobiton"),
          (recruitee, "miaplaza"), (recruitee, "upslide"), (recruitee, "kpsnacks"),
          (personio, "ottonova", "de"), (personio, "pro4all"), (personio, "fact-finder"),
          (bamboohr, "scribd"), (bamboohr, "californiaclosets"), (bamboohr, "mcgregoreba")]
rows = []
for read, *board in BOARDS:
    try:
        rows += read(*board)
    except (OSError, ValueError, KeyError, TypeError, ET.ParseError) as e:
        print("stopped on", board[0], "-", repr(e))

That run gave 35 rows from Workable (6, 29 and 0 from its three boards), 31 from Recruitee (6, 4 and 21), 16 from Personio (8, 4 and 4) and 11 from BambooHR (2, 9 and 0). Every Workable, Recruitee and Personio row had a posted date in the form 2026-11-07, and every BambooHR row had None until bamboohr_date() is called for it. One Recruitee row:

One rowjson
{
  "ats": "recruitee",
  "company": "miaplaza",
  "title": "Spanish Teacher for MiaPrep Online Academy of Ohio (MOAO)",
  "location": "Remote job",
  "remote": true,
  "posted": "2026-11-07",
  "url": "https://miaplaza.recruitee.com/o/spanish-teacher-for-miaprep-online-academy-of-ohio-moao"
}

What it leaves out on purpose: pay (a block of text values on Recruitee, free text on BambooHR, nothing on Workable and Personio), descriptions as clean text, filters, retries and polite pacing (the run waited 1.5 seconds between requests), and the four systems of the earlier post. Workday has a post of its own.

When to stop writing it yourself#

For a handful of boards, the code above is enough. Keep it, and check it when a board changes.

For a list of companies across all these systems, every morning, with the same fields in every row, that is the job of our Greenhouse, Lever, Workday and Ashby jobs scraper on Apify. It reads all eight systems into one shape: Greenhouse, Lever, Workday, Workable, Ashby, Recruitee, Personio and BambooHR. Paste careers page links, up to 200 a run, and it works out the system from the link; a careers page on a company's own domain is not recognised, so write it as recruitee:name, personio:name and so on. It filters by title words, location, remote and days since posting. A row has the description as plain text, and pay where the posting publishes it; a field a system does not publish comes back empty. With description text on, which is the default, it reads each BambooHR posting once more, the extra request this post counted; with it off, it skips that. It reads only pages that the site's robots.txt leaves open, and it is billed per result. What a row holds: every field, one by one.

before you run anything

These are public job postings that employers publish so that people find them. Read them at a gentle pace, keep the link to the original posting, and keep to what the robots.txt of each host leaves open. Every host has its own, and three of these four systems give each company a host of its own: read each before the first request, as the run above did.

written